LAST MILE AI

Against the grain

Omega within the last mile of every Megaport.

Wherever Megaport has a facility, we put a mini Omega data center within the last mile of it — GPUs, CPUs, our own models, our own apps. One hundred terabits per second of dark fiber to the carrier meet. Layer‑2 private enterprise IP to every client. AI that lives in your community, not halfway across the country.

Read the story Pick the hero image

01 · The problem

Today, your AI lives somewhere else.

Every hyperscale model you use sits in a handful of giant campuses. Every question you ask, every call SARAH takes, every document you send, travels to that campus and back — across a state, across a country, sometimes across an ocean. That trip costs three things you never get back: time, energy, and exposure to the public internet.

Packets hauled halfway across the country versus packets that stay local — **#26** Packets: the haul vs local

The long haul is invisible — until you feel it.

A voice turn that should feel instant carries hundreds of milliseconds of pure distance. A conversation over a phone line becomes a relay race between your city and a campus you will never visit. Multiply that by every user, every day, and the “cloud” is quietly the longest commute in your business.

Backhaul across the continent for every request
Energy burned moving bits, not thinking
Latency that no model quality can hide

Energy and latency wasted by centralized hyperscale — **#06** Energy and latency

Centralized is efficient for the landlord. Not for you.

Hyperscale concentrates compute where land and power are cheapest for the operator — and pushes the distance, the latency, and the network bill onto the people who actually use it. The bigger the campus, the further away it is from the community it serves.

We decided to build the opposite.

02 · The idea

One mini Omega beside every Megaport facility.

Megaport already runs a global network of data centers where the carriers meet. Our idea is simple and against the grain: for every Megaport facility, a mini Omega data center within the last‑mile radius. If they are in a city, we are next door. If they have a thousand facilities, we build a thousand mini data centers. One to one.

Megaport facilities paired one to one with Omega mini data centers around the world — **#18** Global pairs — Megaport + Omega

Their footprint becomes our footprint.

We do not pick three regions and ask the world to come to us. We follow the carrier meets — all over the United States, all over the world — and place Omega hyperscale within a mile of each one. Domestic traffic stays domestic. Local traffic stays local.

1:1 — a mini Omega for every Megaport facility
Within the last‑mile radius of the carrier hotel
Same stack in every site — factory, not artisan

The one-mile radius around a Megaport facility served by a mini Omega data center — **#10** The one‑mile radius

The last mile is where the experience is decided.

Whoever sits inside that radius — an enterprise, a hospital, a school, a call center, a city — gets the full 100 terabits per second of GPU speed as if the model were in their own building. It is not a remote service they rent. It is part of the community, close to home.

1 : 1Mini Omega per Megaport facility

100 TbpsDark fiber to each mini DC

< 1 mileFrom the carrier meet

L2 privateEnterprise IP — not the open net

Earth wrapped in filaments of dark fiber connecting Megaport meets to Omega sites

“So whoever is within that radius, within close proximity of that data center, is going to have full access to one hundred terabits per second of GPU speed.”

The founding brief · September 2026

03 · How it connects

Dark fiber in. Private Layer‑2 out.

The mini Omega is joined to the Megaport facility by dark fiber running at 100 terabits per second — enough bandwidth to carry any amount of traffic the community can generate. Clients never touch the public internet to reach us: they connect over Megaport’s Layer‑2 private enterprise IP network, the same way they already reach their carriers.

Layered view of the private path: client site, Megaport Layer-2 network, dark fiber, Omega mini data center — **#20** The private path, layer by layer

No public internet in the data path.

A client’s business never leaves the private fabric. Their site connects to Megaport on Layer‑2. Megaport connects to our racks on dark fiber. The models answer from a mile away. There is no hop through the open internet to expose, intercept, or throttle.

100 Tbps dark fiber — carrier meet to Omega racks
Layer‑2 private enterprise IP — client to Omega
Edge devices at the client for switching and routing
Heavy compute stays in the mini Omega, not on premises

Client premises

A small edge device does the switching and routing. Nothing heavy lives here.

Megaport Layer‑2

The enterprise rides the private network it already trusts — no public internet.

Carrier meet

Where the carriers converge. Our dark fiber begins here.

Mini Omega

Within the last mile. GPUs, CPUs, our models and apps answer from next door.

Dark fiber trench between the Megaport facility and the Omega mini data center — **#24** The fiber trench

Bandwidth is not the constraint any more.

One hundred terabits per second to a site that serves a single metro radius means the pipe is never the bottleneck. Voice, video, documents, screen shares, model calls, connector traffic — all of it, for everyone in the radius, with headroom to spare.

04 · Inside the mini Omega

A real data center. Just a small one, and close.

Each site opens with a minimum of two GB300 racks — Omega hyperscale in miniature — with ribbon racks to follow as the silicon arrives. Alongside the GPUs sit the CPUs, the storage, and the servers that host everything else we run.

Cutaway of a mini Omega data center showing GPU racks, CPUs, storage and fiber entry — **#03** Mini DC cutaway

Everything the community needs, under one small roof.

2× GB300 racks minimum per site — more as demand fills
Ribbon racks when they arrive — the site is built for them
CPUs and servers hosting the apps, connectors, and voice stack
Local storage — data stays in the metro, in the country

Axonometric view of the Megaport facility and Omega site

05 · What runs there

Built by us. Hosted at the last mile.

This is not a colocation cage for someone else’s software. The mini Omega hosts our own large language models, our own speech‑to‑text and text‑to‑speech, and every application we have built on top of them — so the whole experience is domestic, private, and a mile away.

Mosaic of the products hosted in every mini Omega: models, speech, Voice Link, connectors, apps — **#12** The product mosaic

The same sovereign stack in every site.

What lands in one mini Omega lands in all of them. The models, the voices, the phone system, the enterprise connectors, the industry apps — a complete platform, replicated site by site, so a client in one metro gets exactly what a client in another metro gets.

Our own LLMs

Large language models we run ourselves, on GPUs a mile from the people using them — never rented from a distant campus.

Speech — STT and TTS

Streaming speech recognition and natural voices, hosted locally so a conversation feels like a conversation.

Voice Link

A complete telephone system behind one weblink, answered by SARAH — local inference path, community latency.

34.7 million enterprise features

The enterprise connectors, living beside the carriers instead of being hauled across a continent for every call.

25,000+ industry‑specific apps

Applications we developed ourselves, hosted on the same last‑mile fabric as the models that power them.

Our inventions

Everything we have built on top — the conference rooms, the agent consoles, the revenue engine — runs where the client is.

06 · Close to home

AI as part of the community.

The person using the model should feel like they are using a local AI — because they are. It answers from their metro. Their data stays in their country. Their latency is the latency of a mile, not a continent. That is the whole point.

A community — offices, hospital, school, homes — served by a local Omega mini data center — **#04** Community local AI

Everyone in the radius, first class.

An enterprise on Megaport Layer‑2, a hospital, a university, a city hall, a contact center — anyone inside the last‑mile radius gets full local GPU speed. No tiering by distance, no “edge lite”. The full stack, next door.

Latency of a mile — voice that feels present
Data that never leaves the country
Far better experience, because nothing is far

A city with concentric rings of coverage around its Megaport and Omega pair — **#22** City rings

City by city. Country by country.

Because the pattern follows Megaport, it scales the way a network scales — one pair at a time, all over the United States, then all over the world. Every new pair is a new community with its own local AI.

A person on a call with SARAH — the AI feels local because it is

“It’s going to be like they’re using a local AI — because it’s actually part of the community.”

Why we build it this way

07 · The fork in the road

Centralized haul vs last‑mile intelligence

Two ways to deliver the same model. One sends your traffic halfway around the country to a campus that wastes energy and bandwidth on the trip. The other puts the model where the carriers already meet — a mile from you.

Side-by-side: one centralized hyperscale campus versus many local Omega sites — **#02** Centralized vs last mile

Surface-level local site versus a distant hyperscale bunker — **#13** Surface vs bunker

Dimension	Centralized hyperscale	Last Mile AI
Where the model runs	One distant campus, chosen for the operator’s power bill	A mini Omega within a mile of the carrier meet
Distance per request	Across a state, a country, sometimes an ocean	Across a metro
Latency	Hundreds of milliseconds of pure distance	The latency of a mile
Energy	Burned hauling bits back and forth	Spent thinking, not travelling
Network path	Public internet on many hops	Layer‑2 private enterprise IP + dark fiber, end to end
Bandwidth	Shared backbone, contended	100 Tbps dedicated to one radius
Data residency	Wherever the campus happens to be	In your metro, in your country
Whose software	Rented models, rented stack	Our own LLMs, speech, Voice Link, connectors, apps
How it feels	Remote — a service from afar	Local — part of the community

Centralized hyperscale

One campus, continent‑scale backhaul
Traffic travels half the country (or further)
High latency, high energy waste
Public internet exposure for many paths
AI feels remote — rented from afar

Last Mile AI

Mini DC within ~1 mile of each Megaport facility
100 Tbps dark fiber to the carrier hotel
Layer‑2 private enterprise IP end to end
Local GPU speed for everyone in the radius
AI as part of the community — close to home

08 · Why

Against the grain, on purpose.

Against the grain — building local when the industry builds centralized — **#11** Against the grain

The industry is building bigger and further away. We are building smaller and closer.

Hyperscale optimises for the operator. Last Mile AI optimises for the person on the other end of the call. That is why we pair with Megaport instead of competing with it, why we lay dark fiber instead of leasing internet, and why the whole stack is ours to place wherever the community is.

The picture tells it faster than the words: Omega, within the last mile of every Megaport.

Dawn twins — Megaport and Omega side by side